Machine Translation

1. A computer language translation development system, for developing data for use in translation, comprising: means for utilizing corresponding source and target example texts to generate translation units each comprising, in the source and/or target language, a first language item and a second language item of the same language, and means for subsequently generating restrictions on candidate second language items which can be combined with each said first language item.

2. A system according to claim 1 , in which the restrictions comprise for each first language item a list of second language items with which that first language item should not be combined during analysis of source text and/or generation of target text.

3. A computer natural language translation system, arranged to use data developed by a system according to claim 1 .

4. A computer program comprising code to execute on a computer to cause said computer to act as the system of claim 1 .

5. A computer natural language translation system, arranged to use data developed by a system according to claim 1 .

6. A system according to claim 1 , in which said generating means is arranged to select an initial set of restrictions, and to generalize generalise said initial set of restrictions to provide a generalized generalised set of restrictions.

7. A system according to claim 6 , in which said selection of an initial set of restrictions is performed by using said translation units to retranslate said source and target example texts, so as to generate multiple candidate translations thereof, and generating a list of restrictions which, if used to translate said source and target example texts, permit generation of said source and target example texts but limit the number of candidate translations which do not match said source and target example texts.

8. A system according to claim 6 , in which said initial set of restrictions is selected so as to represent, of all possible sets of restrictions, a relatively small number of restrictions.

9. A system according to claim 6 , in which said generating means is arranged to generalize said initial set of restrictions by locating subsets of first language items which exhibit similar restrictions, and by amending the restrictions thereof to increase the similarity of their restrictions.

10. A system according to claim 6 , in which said generalization is performed by induction.

11. A system according to claim 6 , in which said generalization is performed so as to reduce the number of distinct behaviours relative to the number of initial behaviours.

12. A system according to claim 8 , in which said initial set of restrictions is selected so as to represent the smallest possible set of restrictions.

13. A system according to claim 9 , in which the generating means is arranged to allocate each subset of first items a set of restrictions which corresponds to the union of all restrictions of all members of that subset.

14. A computer language translation development system, for developing data for use in translation, comprising: means for generating original translation units each comprising, in the source and/or target language, a first language item, and property data which restricts with which second language items that first language item can occur; and means for generating additional synthetic translation units comprising new combinations of the first language items and property data of said original translation units.

15. A system according to claim 14 , in which said synthetic translation units are generated by: locating translation units which have similar property data.

16. A system according to claim 14 , in which the property data comprises at least one second language item of the same language as the first and with which it must occur in that translation unit.

17. A computer program comprising code to execute on a computer to cause said computer to act as the system of claim 14 .

18. A system according to claim 15 , in which said synthetic translation units are generated by: locating at least first translation unit, and a second translation unit which comprise the same property data as the first with a different first language item to the first; locating a third translation unit which comprises the same first language item as the first translation unit and different property data; and generating a synthetic fourth translation unit which comprises the same first language item as the second translation unit, and the same property data as the third translation unit.

19. A system according to claim 16 , in which the first language item comprises a head of a translation unit and the property data comprises a list of daughters of that head.

20. A method for inferring generalizations of translation units from a set of translation unit examples comprising; the identification of analogous pairs of translation units, each comprising at least one language item and the creation of new translation units which reduce the number of distinct language item behaviours.